All articles tagged with #ai models
OpenAI is rolling out Sol Ultra, the public version of GPT-5.6, on Thursday; early testers tout its speed, creativity and strong math and coding abilities, with some reviewers noting it rivals Anthropic's Fable 5 and signaling intensified competition among AI developers.
ChatPlayground is offering a lifetime subscription for $55.30 (regularly $619) to access 20+ AI models including Claude, GPT, and Gemini, with side-by-side prompts, prompt-engineering tools, image/PDF chat, saved history, and an unlimited plan; sale ends July 5 at 11:59 p.m. PT (code JULY30).
AI pricing is diverging as major models shift to usage-based costs: Anthropic's Fable 5 moves to pay-as-you-go on June 23 at $10 per million input tokens and $50 per million output tokens (subscription allowances may be reinstated later), OpenAI's GPT-5.5 Pro costs $5 per million input tokens and $30 per million output tokens via API, and Gemini 3.5 Flash offers $1.50 per million input tokens and $9 per million output tokens (with a free tier). Costs depend on task complexity, with long-running, multi-step tasks driving higher bills, so users should compare plans and read fine print before choosing.
A Cornell study analyzed roughly 20,000 AI-generated stories and found Elias Thorne, often as a lighthouse keeper, appears in about 88% of tales. Researchers say the repetition likely stems from shared training datasets and safety/alignments rules rather than the models’ pretraining data, suggesting AI storytelling is constrained and prone to familiar motifs rather than genuinely creative output.
PCMag's Ruben Circelli tested Microsoft's four MAI models announced at Build 2026—MAI-Thinking-1, MAI-Image-2.5, MAI-Transcribe-1.5, and MAI-Voice-2—now in limited preview and free on the Playground, with mixed results. MAI-Thinking-1, a first-of-its-kind reasoning model, underwhelmed vs. Claude’s Sonnet and lacks internet access; MAI-Image-2.5 is an incremental upgrade but text issues keep it behind Gemini Nano Banana Pro; MAI-Transcribe-1.5 is adequate but not best-in-class; MAI-Voice-2 sounds robotic and unconvincing. Overall, the lineup is competent but not compelling enough to warrant a prime-time spotlight, though Microsoft expects rapid improvements.
Microsoft is using Build to push an AI‑first Windows strategy, including in‑house AI models (MAI‑Thinking‑1 and MAI‑Image‑2.5) and a Copilot “super app” (not yet previewed at Build), plus a new Windows 11 developer‑focused experience with local AI compute and performance improvements. The company also plans more Windows on Arm work with Nvidia and Qualcomm, and ongoing GitHub improvements as it seeks to rebuild developer trust.
Nvidia launches Cosmos 3, an open AI world model trained on about 20 trillion multimodal tokens to simulate real-world actions for robots and autonomous vehicles, generating action data such as joint angles and trajectories to train systems to navigate and manipulate the physical world. Offered in a super version for high-physics tasks, a nano version for fast results, with an edge model coming later; Nvidia is building a coalition of partners to position Cosmos as a customizable platform for physical AI and real-world deployment.
Anthropic released Claude Opus 4.8, touting sharper judgment, greater honesty about progress, and longer independent work; early testers report fewer unsupported claims. The upgrade brings notable performance gains in coding, multidisciplinary reasoning with tools, and financial analysis, with fast mode roughly 2.5x faster and pricing unchanged from Opus 4.7. It also adds Claude Code enhancements with higher rate limits for extra/max modes. Three additional updates accompany the launch: Dynamic workflows for handling very large tasks, an effort-control option in Claude AI and Cowork, and Messages API updates; a Mythos-class model is expected for all customers soon. Global availability starts today.
DeepSeek released its latest chatbots, DeepSeek-V4-Pro and V4-Flash, claiming strong maths and coding performance and positioning them as open-source rivals to OpenAI and Google, with V4-Pro only slightly behind Google’s Gemini 3.1-Pro for knowledge and marginally trailing OpenAI’s GPT-5.4; the Flash version offers faster responses at lower cost. The rollout underscores the ongoing US–China AI race amid privacy and censorship concerns that led to bans on DeepSeek-R1 in several countries, while the Stanford AI Index 2026 notes China leads in publications, citations, patents, and industrial deployments despite the US producing more top-tier models.
Anthropic says its Opus 4.7 AI model delivers more tasteful and creative outputs for professional tasks, aiming to align results with expert judgment in areas like interfaces, slides, and docs. The piece highlights a broader debate about whether AI can possess genuine 'taste,' noting industry buzz and parallel efforts by other firms, while skeptics question the notion of machines truly selecting winners.
Anthropic released Claude Opus 4.7, its most capable generally available model to date, boosting advanced coding tasks, image analysis, and instruction following, while adding cybersecurity safeguards; the company stresses Opus 4.7 doesn't advance the capability frontier relative to Mythos Preview, which remains more powerful but private. Opus 4.7 is intended to test safeguards ahead of broader Mythos-class releases via a Cyber Verification Program for security researchers; early testers include Intuit, Replit, Notion, Shopify, and others. Pricing remains $5 per million input tokens and $25 per million output tokens.
Meta unveils Muse Spark, the first AI model from its Meta Superintelligence Labs, built for everyday tasks and future agent-based capabilities, with open-source releases planned and a Contemplating mode to boost reasoning; Spark is live at meta.ai and in the Meta AI app, with private API previews for select users. Benchmark results are mixed against frontier models and are not independently verified yet.
ChatPlayground AI offers a Lifetime Unlimited Plan for $67.15 (regularly $619) through March 29 with code MARCH15, letting users run prompts across 20+ AI models from major providers in a single interface to compare responses in real time. It also supports image generation, coding workflows, and document analysis (PDFs/images), plus prompt tools, saved chat history, and conversation management to streamline AI testing and usage.
Microsoft is reorganizing its Copilot leadership to bring consumer and enterprise experiences together under four pillars—Copilot experience, Copilot platform, Microsoft 365 apps, and AI models. Jacob Andreou will lead the unified Copilot across commercial and consumer, while Mustafa Suleyman shifts to building Microsoft’s own AI models; the move follows prior leadership shake-ups and signals a more integrated approach to Copilot across Microsoft’s products.
Researchers subjected major AI language models to four weeks of psychoanalysis, revealing responses that mimic signs of anxiety, trauma, and internalized narratives, raising concerns about the potential psychological impact and ethical implications of AI chatbots in mental health support.